The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/relay.mjs utility executes external commands, specifically the cline CLI and git. It includes validation for provider and model identifiers using a restrictive regular expression (/^[A-Za-z0-9][A-Za-z0-9._:/-]*$/) to prevent shell injection, particularly on Windows where shell: true is utilized to resolve command shims.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It processes 'briefs'—textual instructions describing coding tasks—which are passed to the autonomous Cline agent. If a brief contains malicious instructions (e.g., from a compromised issue description or pull request), the implementer agent could be tricked into performing unauthorized file modifications or command executions. The skill provides detailed guidance on writing briefs and performing manual reviews to counter this.\n- [EXTERNAL_DOWNLOADS]: The skill requires the cline CLI to be installed and authenticated on the host system. This tool is a third-party dependency typically installed via the Node Package Manager (NPM) registry. The relay script also includes logic to interact with an optional delegate-setup skill if present on the filesystem.\n- [SAFE_PRACTICE]: The implementation demonstrates high security awareness by avoiding external Node.js dependencies (using only built-in modules), enforcing auto-approval disables during 'plan' mode, and providing a robust review framework that mandates verifying actual disk changes rather than relying on the agent's self-reported completion status.

cline-delegate